MIRAGE anchors feed-forward vision to LiDAR-IMU motion for metric SLAM and Gaussian reconstruction
zhenjun_zhao · x · 2026-10-13
The arXiv paper MIRAGE anchors feed-forward visual geometry priors to LiDAR–IMU metric motion, jointly solving the missing metric scale and long-horizon drift of vision-only reconstruction.
Method highlights:
- Local: metric-visual pose refinement with keyframe GN, LiDAR-IMU BA, and point-plane constraints
- Global: reliability-aware loop correction where correction magnitude is gated by correspondence support and cross-modal agreement
- Streaming online Gaussian mapping plus offline geometry-aware refinement (Split/Replace/Insert driven by RGB residuals + LiDAR support)
Results: accurate metric trajectories on Oxford Spires, M2DGR and INS with 0.73s online latency, outperforming SLAM baselines and beating COLMAP offline; rendering quality (PSNR/SSIM/LPIPS) leads Gaussian-LIC2 on most sequences. The authors also built a multimodal data-collection platform and released a new vision-only long-horizon indoor benchmark.
More from Research
- Allen AI's TutorMoments Benchmark Finds LLM Tutors Over-Help Students — allen_ai · 2026-10-13
- Ben Goertzel revisits Omohundro's Basic AI Drives, finding six assumptions that decide which minds apply — bengoertzel · 2026-10-13
- Sim-to-real skepticism: pixel-space native training beats generated simulators, argues DeWitte — chris_j_paxton · 2026-10-13
- DeepMind publishes survey mapping theories of machine consciousness — Dr_Atoosa · 2026-10-13
- Noise: a speedrun for pretraining transformers without backprop — madhavsinghal_ · 2026-10-13
- Harvard releases freeinference_agentic_trace, a million-scale agent trace dataset — harvardMadsys · 2026-10-13